bzr commit into mysql-5.1-telco-6.3 branch (magnus.blaudd:2993) Bug#45899

[Magnus Blåudd]

#At file:///home/msvensson/mysql/6.3-bug45899/ based on revid:jonas@strippedrqfv8drfi1wk7

 2993 Magnus Blåudd	2009-07-03
      Bug#45899 api nodes (re-)connecting to starting data nodes too fast ...
       - The ndbd starts to accept incoming connections from other nodes quite early
         in startup, initially only ndbd and mgm nodes are allowed to connect. During
         this time all api nodes will quickly be disconnected, since the api nodes
         will retry connect quite rapidly this may exhaust the available ports in the
         machine.
       - Since all incoming connections are handled by a separate thread, we can
         allow it to sleep and wait for a little while. Hopefully the incoming connection
         will be allowed after the 300*300ms pool loop. But even if that is not the case
         the rate of connection attempts have been reduced.

    modified:
      storage/ndb/include/transporter/TransporterRegistry.hpp
      storage/ndb/src/common/transporter/TransporterRegistry.cpp
=== modified file 'storage/ndb/include/transporter/TransporterRegistry.hpp'
--- a/storage/ndb/include/transporter/TransporterRegistry.hpp	2009-05-27 12:11:46 +0000
+++ b/storage/ndb/include/transporter/TransporterRegistry.hpp	2009-07-03 10:04:32 +0000
@@ -370,6 +370,8 @@ private:
 
   int m_shm_own_pid;
   int m_transp_count;
+
+  bool allow_wait_connecting() const;
 };
 
 inline void

=== modified file 'storage/ndb/src/common/transporter/TransporterRegistry.cpp'
--- a/storage/ndb/src/common/transporter/TransporterRegistry.cpp	2009-05-27 12:11:46 +0000
+++ b/storage/ndb/src/common/transporter/TransporterRegistry.cpp	2009-07-03 10:04:32 +0000
@@ -217,6 +217,19 @@ TransporterRegistry::init(NodeId nodeId)
 }
 
 bool
+TransporterRegistry::allow_wait_connecting() const {
+  DBUG_ENTER("TransporterRegistry::allow_wait_connecting");
+
+  // Check if any node is in connecting
+  for (unsigned i=0; i<maxTransporters; i++) {
+    if (performStates[i] == CONNECTING)
+      DBUG_RETURN(false);
+  }
+
+  DBUG_RETURN(true);
+}
+
+bool
 TransporterRegistry::connect_server(NDB_SOCKET_TYPE sockfd)
 {
   DBUG_ENTER("TransporterRegistry::connect_server");
@@ -256,9 +269,20 @@ TransporterRegistry::connect_server(NDB_
       DBUG_RETURN(false);
   }
 
-  //check that the transporter should be connected
-  if (performStates[nodeId] != TransporterRegistry::CONNECTING) {
-    DBUG_PRINT("error", ("Transporter in wrong state for this node id from client"));
+  // check that the transporter should be connected
+  int loops = 0;
+  while (performStates[nodeId] != TransporterRegistry::CONNECTING) {
+    DBUG_PRINT("info", ("Transporter[%d] is not in CONNECTING", nodeId));
+    if (allow_wait_connecting() && loops++ < 300)
+    {
+      DBUG_PRINT("info", ("Allowing connection to wait for CONNECTING"));
+      g_eventLogger->info("Connection from node %d is waiting for CONNECTING",
+                          nodeId);
+      NdbSleep_MilliSleep(300);
+      continue;
+    }
+
+    DBUG_PRINT("error", ("Disconnecting node %d, not CONNECTING"));
     DBUG_RETURN(false);
   }
 

Attachment: [text/bzr-bundle] bzr/[e-mail]